Re: Information on ioctl32

[James Hansen]

Arnd Bergmann wrote:

> On Maandag 24 Oktober 2005 12:48, James Hansen wrote:
>  
>
> > Sorry, I'll explain a little more about I'm doing.
> >    
>
> Btw, please don't top-post.
>
>  
Whoops..  sorry :(

> > The kernel I'm running is the default kernel from the (unofficial) 
> > debian amd64 distro.  uname says it's 2.6.8-11-amd64-generic.  When 
> > running a 32bit app, linked against the 32bit libraries, calling a 64bit 
> > driver in, the kernel logs messages saying something along the lines of 
> > "ioctl32 not defined".
> >    
>
> Ok, that is a little clearer then. It's not missing the infrastructure
> but rather a conversion for a specific ioctl number in that device
> driver. Actually, the message should print out which number that is,
> otherwise, you can probably find out by running the program with
> 'strace' and looking for ioctl() calls that result in 'EINVAL'
> or 'ENOTTY'.
>
>  
I should probably mention that this is a 3rd party device driver for a 
PCI card, which is not part of the kernel itself.  I'm just looking to 
add 32bit compatibility options to make this device driver work with as 
many kernel versions as possible.

> > The headers I'm building my modules against are from the package 
> > 'kernel-headers-2.6.8-11-amd64-generic' which leaves me with the 
> > directory 'kernel-headers-2.6.8-11-amd64-generic' in /usr/src. 
> >
> > My problem is that when I look in the headers at the file_operations 
> > struct for compat_ioctl there's no entry there, and I therefore can't 
> > define a function for it.  I had no idea there was a dynamic system in 
> > place before what's mentioned on lwn.  I'd assumed that as my kernel was 
> > trying to call ioctl32, that it would have had the patch applied, and 
> > it's headers should have contained an appropriate entry in file_operations.
> >
> > So it looks like I'll have implement both ways of doing things, one for 
> > pre-2.6.11 and another for post 2.6.11 kernels.
> >
> > Would you know of anywhere else I could look for information on the 
> > dynamic method you mentioned that existed in kernels before 2.6.11.
> >    
>
> Actually, you should not implement anything for old kernels, they are
> just not supported. The only thing that helps is that you find out
> which driver is missing the conversion and (find someone to) add the
> compat_ioctl method to the device driver for 2.6.15.
>
> The interesting information here is
> - name of the device driver
> - name of the application
> - all the ioctl numbers that are missing (can be found out from
>  the other two)
>
> If you really want to do the extra work to get it running on kernels
> older than 2.6.11, look for register_ioctl32_conversion(). Just know
> that we had a good reason to replace that and I don't think you will
> have much lock in getting the handler for your driver into the
> Debian 2.6.8 kernel.
>
> 	Arnd <><
>  
I'll put in a check for kernel version, and use compat_ioctl in the 
file_operations struct for 2.6.11.  And then I'll have a read about 
register_ioctl32_conversion for versions of the kernels pre-2.6.11.

I'll also have a read about why it was changed, but just out of interest :)

Thanks Arnd,

James

> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
>
> .
>
>  

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/